In [5]:
import tensorflow_datasets as tfds
import tensorflow as tf
(raw_train, raw_validation, raw_test), metadata = tfds.load(
'tf_flowers',
split=['train[:80%]', 'train[80%:90%]', 'train[90%:]'],
with_info=True,
as_supervised=True,
)
In [6]:
import matplotlib.pyplot as plt
%matplotlib inline
%config InlineBackend.figure_format = 'retina'
In [7]:
plt.figure(figsize=(10, 5))
get_label_name = metadata.features['label'].int2str
for idx, (image, label) in enumerate(raw_train.take(10)):
plt.subplot(2, 5, idx+1)
plt.imshow(image)
plt.title(f'label {label}: {get_label_name(label)}')
plt.axis('off')
In [8]:
IMG_SIZE = 160 # All images will be resized to 160x160
def format_example(image, label):
image = tf.cast(image, tf.float32) # image=float(image)같은 타입캐스팅의 텐서플로우 버전입니다.
image = (image/127.5) - 1
image = tf.image.resize(image, (IMG_SIZE, IMG_SIZE))
return image, label
In [9]:
train = raw_train.map(format_example)
validation = raw_validation.map(format_example)
test = raw_test.map(format_example)
In [31]:
plt.figure(figsize=(10, 5))
get_label_name = metadata.features['label'].int2str
for idx, (image, label) in enumerate(train.take(10)):
plt.subplot(2, 5, idx+1)
image = (image + 1) / 2
plt.imshow(image)
plt.title(f'label {label}: {get_label_name(label)}')
plt.axis('off')
In [32]:
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense, Conv2D, Flatten, MaxPooling2D
In [42]:
model = Sequential([
Conv2D(filters=16, kernel_size=3, padding='same', activation='relu', input_shape=(160, 160, 3)),
MaxPooling2D(),
Conv2D(filters=32, kernel_size=3, padding='same', activation='relu'),
MaxPooling2D(),
Conv2D(filters=64, kernel_size=3, padding='same', activation='relu'),
MaxPooling2D(),
Flatten(),
Dense(units=512, activation='relu'),
Dense(units=5, activation='softmax')
])
In [43]:
import numpy as np
image = np.array([[1, 2], [3, 4]])
print(image.shape)
image
Out[43]:
In [44]:
image.flatten()
Out[44]:
In [45]:
base_learning_rate = 0.0001
model.compile(optimizer=tf.keras.optimizers.RMSprop(lr=base_learning_rate),
loss=tf.keras.losses.sparse_categorical_crossentropy,
metrics=['accuracy'])
In [46]:
BATCH_SIZE = 32
SHUFFLE_BUFFER_SIZE = 1000
In [47]:
train_batches = train.shuffle(SHUFFLE_BUFFER_SIZE).batch(BATCH_SIZE)
validation_batches = validation.batch(BATCH_SIZE)
test_batches = test.batch(BATCH_SIZE)
In [48]:
for image_batch, label_batch in train_batches.take(1):
pass
image_batch.shape
#print(train_batches)
#print(validation_batches)
Out[48]:
In [49]:
validation_steps = 3
loss0, accuracy0 = model.evaluate(validation_batches, steps=validation_steps)
print("initial loss: {:.4f}".format(loss0))
print("initial accuracy: {:.4f}".format(accuracy0))
In [50]:
EPOCHS = 10
history = model.fit(train_batches,
epochs=EPOCHS,
validation_data=validation_batches)
In [51]:
acc = history.history['accuracy']
val_acc = history.history['val_accuracy']
loss=history.history['loss']
val_loss=history.history['val_loss']
epochs_range = range(EPOCHS)
plt.figure(figsize=(4, 4))
plt.subplot(1, 1, 1)
plt.plot(epochs_range, acc, label='Training Accuracy')
plt.plot(epochs_range, val_acc, label='Validation Accuracy')
plt.legend(loc='lower right')
plt.title('Training and Validation Accuracy')
#plt.subplot(1, 2, 2)
plt.plot(epochs_range, loss, label='Training Loss')
plt.plot(epochs_range, val_loss, label='Validation Loss')
plt.legend(loc='upper right')
plt.title('Training and Validation Loss')
plt.show()
In [52]:
for image_batch, label_batch in test_batches.take(1):
images = image_batch
labels = label_batch
predictions = model.predict(image_batch)
pass
predictions
Out[52]:
In [53]:
import numpy as np
predictions = np.argmax(predictions, axis=1)
predictions
Out[53]:
In [54]:
plt.figure(figsize=(20, 12))
for idx, (image, label, prediction) in enumerate(zip(images, labels, predictions)):
plt.subplot(4, 8, idx+1)
image = (image + 1) / 2
plt.imshow(image)
correct = label == prediction
title = f'real: {label} / pred :{prediction}\n {correct}!'
if not correct:
plt.title(title, fontdict={'color': 'red'})
else:
plt.title(title, fontdict={'color': 'blue'})
plt.axis('off')
In [55]:
count = 0 # 정답을 맞춘 개수
for image, label, prediction in zip(images, labels, predictions):
image = (image + 1) / 2
correct = (label == prediction)
if correct:
count = count + 1
print(count / 32 * 100, '%')
In [56]:
IMG_SHAPE = (IMG_SIZE, IMG_SIZE, 3)
# Create the base model from the pre-trained model VGG16
base_model = tf.keras.applications.VGG16(input_shape=IMG_SHAPE,
include_top=False,
weights='imagenet')
In [57]:
image_batch.shape
Out[57]:
In [58]:
feature_batch = base_model(image_batch)
feature_batch.shape
Out[58]:
In [59]:
base_model.summary()
In [60]:
feature_batch.shape
Out[60]:
In [61]:
import numpy as np
image = np.array([[1, 2],
[3, 4]])
flattened_image = image.flatten()
print("Original image:\n", image)
print("Original image shape:", image.shape)
print()
print("Flattened image:\n", flattened_image)
print("Flattened image shape:", flattened_image.shape)
In [62]:
global_average_layer = tf.keras.layers.GlobalAveragePooling2D()
In [63]:
feature_batch_average = global_average_layer(feature_batch)
print(feature_batch_average.shape)
In [82]:
dense_layer = tf.keras.layers.Dense(512, activation='relu')
prediction_layer = tf.keras.layers.Dense(5, activation='softmax')
# feature_batch_averag가 dense_layer를 거친 결과가 다시 prediction_layer를 거치게 되면
prediction_batch = prediction_layer(dense_layer(feature_batch_average))
print(prediction_batch.shape)
In [83]:
base_model.trainable = False
In [84]:
model = tf.keras.Sequential([
base_model,
global_average_layer,
dense_layer,
prediction_layer
])
In [85]:
model.summary()
In [86]:
base_learning_rate = 0.0001
model.compile(optimizer=tf.keras.optimizers.RMSprop(lr=base_learning_rate),
loss=tf.keras.losses.sparse_categorical_crossentropy,
metrics=['accuracy'])
In [88]:
validation_steps=10
loss0, accuracy0 = model.evaluate(validation_batches, steps = validation_steps)
print("initial loss: {:.2f}".format(loss0))
print("initial accuracy: {:.2f}".format(accuracy0))
In [111]:
EPOCHS = 20 # 이번에는 이전보다 훨씬 빠르게 수렴되므로 5Epoch이면 충분합니다.
history = model.fit(train_batches,
epochs=EPOCHS,
validation_data=validation_batches)
In [112]:
acc = history.history['accuracy']
val_acc = history.history['val_accuracy']
loss = history.history['loss']
val_loss = history.history['val_loss']
plt.figure(figsize=(8, 4))
plt.subplot(1, 2, 1)
plt.plot(acc, label='Training Accuracy')
plt.plot(val_acc, label='Validation Accuracy')
plt.legend(loc='lower right')
plt.ylabel('Accuracy')
plt.ylim([min(plt.ylim()),1])
plt.title('Training and Validation Accuracy')
plt.subplot(1, 2, 2)
plt.plot(loss, label='Training Loss')
plt.plot(val_loss, label='Validation Loss')
plt.legend(loc='upper right')
plt.ylabel('Cross Entropy')
plt.ylim([0,1.0])
plt.title('Training and Validation Loss')
plt.xlabel('epoch')
plt.show()
In [113]:
for image_batch, label_batch in test_batches.take(1):
images = image_batch
labels = label_batch
predictions = model.predict(image_batch)
pass
predictions
Out[113]:
In [114]:
import numpy as np
predictions = np.argmax(predictions, axis=1)
predictions
Out[114]:
In [115]:
plt.figure(figsize=(20, 12))
for idx, (image, label, prediction) in enumerate(zip(images, labels, predictions)):
plt.subplot(4, 8, idx+1)
image = (image + 1) / 2
plt.imshow(image)
correct = label == prediction
title = f'real: {label} / pred :{prediction}\n {correct}!'
if not correct:
plt.title(title, fontdict={'color': 'red'})
else:
plt.title(title, fontdict={'color': 'blue'})
plt.axis('off')
In [116]:
count = 0
for image, label, prediction in zip(images, labels, predictions):
correct = label == prediction
if correct:
count = count + 1
print(count / 32 * 100) # 약 95% 내외
In [ ]: